How do you know whether a hedge actually works?

A hedge can only be judged against a specific problem. If you say a hedge is there to reduce BTC beta, that is different from reducing alt-sector beta, event gap risk, or tail volatility. Vague goals produce vague evaluations.

The cleanest checks ask whether the hedged trade behaves better than the unhedged trade on the exact dimension you wanted to control.

Hedge relationships often look more stable in historical data than they do in live trading. Correlations shift, basis widens, liquidity thins out, and one leg may be much harder to execute at size than the other.

That is why hedge quality should be checked in rolling windows and then rehearsed in paper trading. A hedge that only works under static assumptions is usually not a robust hedge.

Treat hedge quality as a comparison problem. Run the unhedged and hedged versions with the same cost model, track residual exposure over time, and keep the hedge rule attached to the run metadata so you can tell whether the improvement came from real risk removal or from accidental fitting.